ZILBER’S PSEUDO-EXPONENTIAL FIELDS

WILL BONEY

The goal of this talk is to prove the following facts above Zilber’s pseudoexpo- nential fields

(1) they are axiomatized in Lω1,ω(Q) and this logic is essential. (2) they form a quasiminimal class ∼ (3) C ≡ B implies C = B The results and presentation are drawn from Kirby “A note on the axioms for Zilber’s Pseudo-Exponential Fields” [Kir] and Bays and Kirby “Excellence and uncountable categoricity of Zilber’s Exponential Fields” [BaKi]. ∗ The class of pseudo-exponential fields is denoted ECst,ccp and is the collection of structures hF ; +, ·, expi satisfying the following axioms: · (1) ELA-field: F models ACF0 and exp : F → F is a surjective homomor- phism (2) Standard kernel: ker exp is an infinite cyclic group generated by a tran- scendental element 2πi. (3) Schanuel property: For all x ∈ F ,

δ(x) := td(x, exp(x)) − ldimQ(x) ≥ 0 (4) Strong exponential-algebraic closedness: Every system of exponential has a solution, unless this would violate Schanuel. (5) Countable Closure Property: eclF is a pregeometry such that, if C ⊂ F is finite, then eclF (C) is countable. The fourth axiom is imprecisely stated, but we will state it more precisely. We will change the language (essentially but conservatively) to show this is a quasiminimal class. First, we define some useful notions.

δ(x/A) := td(x, exp x, A, exp A/A, exp A) − ldimQ(x/A) Second, we discuss the axioms. (1) ELA-field: Clear.

Date: August 12, 2015. This material is based upon work done while the first author was supported by the National Science Foundation under Grant No. DMS-1402191. 1 2 WILL BONEY

(2) Standard kernel: This is Lω1,ω expressible. The important bit is that this defines the integers: Z(F ) := {r ∈ F : ∀x[x ∈ ker exp → rx ∈ ker exp]} An alternate approach would to just say that Z(F ) is standard (using

Lω1,ω); then the standard kernel is first-order. (3) Schanuel property: This is expressible as a first order scheme, modulo n n the integers being standard: for each variety V ⊂ Ga × Gm defined over Q of dimension n − 1, n ! X ∀x∃m ∈ Z(F ) − {0} (x, exp(x)) ∈ V → mixi = 0 i=1 Essentially, this says that any tuple with td(x, exp(x)) < n has some linear dependence. This property is a big part of the interest in Zilber’s work. Schanuel’s Conjecture says that C has satisfies this and, not to understate it’s im- portance, Wikipedia1 says something like proving Schanuel’s Conjecture “would generalize most known results in transcendental number theory.” (4) Strong exponential-algebraic closedness: Before we state this for- mally, we need some definitions. Write G for Ga × Gm.

n Given a matrix M ∈ Matn×n(Z), this acts on G by being an additive n n map on Ga and a multiplicative map on Gm. Write G · V for the image of V ⊂ Gn under M. An irreducible variety V ⊂ Gn is rotund iff for every M ∈ Matn×n(Z), dimM · V ≥ rkM. Let (x, y) be a generic point of V over F . V is additively free iff the xi don’t satisfy X mixi ∈ F i

mi Πi

for any mi ∈ Z, not all 0.

The property is: n n If V is a rotund, additively and multiplicatively free subvariety of Ga ×Gm defined over F and of dimension n and a is a finite tuple from F , then there is x ∈ F such that (x, exp x) ∈ V is generic in V over a. So if there are some exponential polynomials that are nice enough, then there is a root.

1http://en.wikipedia.org/wiki/Schanuel’s_conjecture#Consequences ZILBER’S PSEUDO-EXPONENTIAL FIELDS 3

Proposition 0.1 ( [Kir]). This is first-order expressible modulo the previ- ous axioms.

Proof: The key is that we have Z already definable. Suppose we have n a parametric family of varieties (Vp)p∈P from G , where P is the parame- terizing variety. It is known (fibre dimension theory) that {p ∈ P : Vp is irreducible of dimension n} is first order definable. 2 With Z , we can define additive freeness: ! X ∀m ∈ Z − {0}∀z∃x x ∈ Vp ∧ mixi 6= z i

n Given a parametric family of varieties (Vp)p∈P from G , where P is the 0 parameterizing variety, let P ⊂ P be the subset so Vp is irreducible, of dimension n, rotund, and additively and multiplicatively free. Then we add the following scheme 0 r n ∀p ∈ P ∀a ∈ F ∃x ∈ F ∀m ∈ Q " !# X X (x, exp x) ∈ Vp ∧ mixi + mn+iai = 0 → ∧i

2 Necessary; x1 + px2 = 0 is additively free iff p 6∈ Q 3Note that xy is definable as exp(y log x), which is well-defined for integer y. 4Parametric varieties are the equivalent of fixing the degree and quantifying over all coefficients in algebraically closed fields. 5To do this, let a0 be a minimizer of δ(y/a) as y ranges over F . Then we have δ(y/aa0) = δ(yaa0) − δ(aa0) ≥ 0 (by choice of a0), so use aa0 in place of a. 4 WILL BONEY

The scheme then gives a x such that ldimQ(x/a) = n. By Schanuel, then td(x, exp x/a, exp a) ≥ n. Since V is of dimension n, we have (x, exp x) is generic in V over (a, exp a), which is more than we need. †

[Kir] says that this was originally described by [Zil05] as a “slight sat- uratedness” of the structure, so it being first order was a bit of a surprise. Having it first order allows for the some of the details to be simplified. On the other hand, the analogy to algebraic closure makes it being first order more expected. (5) Countable Closure Property: There are two definitions of exponential closure that are the same with axiom 2 (we will not prove this, see [Kir10, Theorem 1.3]). We define the one used in [Kir]. An exponential f(X) = p(X, exp(X)), where p ∈ F [X, Y]. A Khovanskii system of width n is exponential polynomials f0, . . . , fn−1 of arity n such that

fi(x) = 0∀i < n ∂f0 ∂f0 ... ∂x0 ∂xn−1 ...... (x) 6= 0 ∂fn−1 ... ∂fn−1 ∂x0 ∂xn−1 Then F ecl (C) := {a0 ∈ F : a ∈ F is the solution to the Khovanski system given by f0, . . . , fn−1 defined over Q(C)} By [Kir10, Theorem 1.1], this is a pregeometry in any exponential field. We also have an easy proof that this is L(Q).

Proposition 0.2 ( [Kir]). Axiom 5 is an L(Q) scheme.

Proof: Given exponential polynomials , let χ(x, z) state that x is a solution to the Khovanskii system with coefficients z.

∀z¬Qx0∃x1, . . . , xn−1χf (x, z) † We’re going to discuss two interesting things coming from [Kir] and then look at the language used for quasiminimality.

Q cannot be eliminated: We’re going a little out of order here, because we will use some of the quasiminimality in this proof.

Proposition 0.3 ( [Kir]). Countable closure is not L∞,ω expressible. ZILBER’S PSEUDO-EXPONENTIAL FIELDS 5

Proof: Let F0 be the zero dimensional Zilber field. Then we can adjoin some a 6 so exp a = a and form the strong exponential algebraic closure to get F1 ∼ . Note that F1 is also a zero dimensional Zilber field, so F0 = F1. Our goal is to show that F0 ≺L∞,ω F1. Let b ∈ F0 and let B be the smallest ELA-subfield containg b. By general Fraisse construction or [Kir11, Proposition 6.9], both F0 and F1 are isomorphic to ∼ the strong exponential algebraic closure of b. That is F0 =B F1. Thus, given any L∞,ω(b) formula, F0 and F1 agree on it. We iterate this process to form {Fα : α < ω1}. Note at limit stages, L∞,ω substructure is well-behaved (L(Q) is not) and taking unions cannot grow the ex- ponential transcendence degree. Thus Fω1 is L∞,ω equivalent to all of these, but does not satisfy the countable closure property. †

Our restriction to L∞,ω formulas comparing F0 and F1 was not crucial; this is true of any logic. However, the key step is that ≺L∞,ω is stable under unions of chains of any (esp. uncountable) length. Note that this gives the first7 explicit example of a non-finitary AEC.

Elementary equivalence implies categoricity: Again, we go out of order. This is [Kir, §2.5]. The key is that all of the axioms except 2 and 5 are first order (modulo 2). Obviously, C has standard Z, so we only need to see that countable closure property holds. [Zil05, Lemma 5.12] proves this, but the Khovanskii-system definition of ecl gives a shorter proof. Suppose f0, . . . , fn−1 is a Khovanskii system with coefficients from Q(C). The inequation in the system says that the Jacobian of f0, . . . , fn−1 is nonzero. Then we can use the implicit theorem to find an open set around each solution to the Kohvanskii system such that it is the only solution in that set: Let f(y, x) = (f0(x), . . . , fn−1(x)) and note f(0, a) = 0. By the implicit function theorem, there is • open U 3 0 • open Va 3 a • g : U → V such that

{(x, y) ∈ U × Va : f(x, y) = 0} = {(x, g(x)) : x ∈ U}

The key point is that the only solution in Va is a. Thus, the solutions are isolated in the complex topology, and there are only countably many of them.

6 In doing this, we need to ensure that F1 has no nonstandard kernel elements. First, in the q Q-linear span of F0 and a, we have that exp(f + qa) = 1 iff exp(f)a = 1 iff exp(f) = 1 and q = 0. So no new kernel elements are in this span. To get F1, we form the ELA-closure and then the strong exponential algebraic closure of this span, each of which does not add new kernel elements. More details are in Kirby [Kir11]. 7Potentially not the first chronologically, but at least the first ones that people got excited about. 6 WILL BONEY ∼ Thus, C is the Zilber field of dimension continuum. So, C = B. †

The language of quasiminimality: One facet of quasiminimal classes is that quantifier free types must be the right one, i. e., Galois type. This is not true in r+n r+n the current language, so we expand it. For each algebraic variety V ⊂ Ga ×Gm that is defined and irreducible over Q, we set the r-ary relation r+n RV,n(x) := ∃y∀m ∈ Q ((x, y, exp(x), exp(y)) ∈ V ∧ ! X X mixi + mi+ryi = 0 → ∧r≤i

Set L = h+, 0, λ·,RV,niλ∈Q. Note the similarity between these relations and the statement of Axiom 4

0 r n ∀p ∈ P ∀a ∈ F ∃x ∈ F ∀m ∈ Q " !# X X (x, exp x) ∈ Vp ∧ mixi + mn+iai = 0 → ∧i

If (Vp)p∈P is a parameterizing variety in (x, exp x) and (Wp)p∈P is the parame- terizing variety in (x, y, exp x, exp y) requiring only the same relationship between y and exp y, then an instance of Axiom 4 is equivalent to 0 r ∀p ∈ P ∀a ∈ F RWp,n(a) The reason we change the language is the following lemma of Zilber: Fact 0.4 ( [Zil05].5.7). Suppose A ≺ B and a, b are finite tuples so Aa ≺ B and ∼ 8 Ab ≺ B. If there is an isomorphism so FAa =FA FAb that sends a to b, then tpqf,L(a/A) = tpqf,L(b/A). Additionally, making the language more relational means that structures can be smaller. We call this change conservative since L has the same definable sets as the original language (in some Zilber field): First, note that V = V (x1x2 = x3) and RV,0 define the graph of multiplication:

RV,0(a1, a2, a3) ⇐⇒ a1a2 = a3 0 and W = V (x2 = x1) and RW,0 define the graph of :

RV,0(a1, a2) ⇐⇒ a2 = exp(a1) ∗ Second, Q and, thus, each RV,n is definable in each member of ECst,ccp.

8 FX is the field generated by X ∪ exp(X) with exp regarded as a partial function defined only on X. Note that this is not a structure in the original language, but is a structure in the new one. ZILBER’S PSEUDO-EXPONENTIAL FIELDS 7

Zilber initially proves that this is a quasiminimal class in [Zil05], but Bays and Kirby fill some gaps/make some corrections. Note, following [BHH+12], proving excellence is not necessary.

References [BHH+12] Martin Bays, Bradd Hart, Tapani Hyttinen, Meeri Kesala, and Jonathan Kirby, Quasiminimal structures and excellence. Bulletin of the London Mathematical Society 46(1), p. 155-163, 2014. [BaKi] Martin Bays and Jonathan Kirby, Excellence and Categoricity of Zilber’s Exponential Fields [Kir] Jonathan Kirby, A Note on the Axioms for Zilber’s Pseudo-Exponential Fields. Notre Dame Jounal of Formal Logic 54(3-4), p. 509-520, 2013. [Kir10] Jonathan Kirby, Exponential Algebraicity in Exponential Fields. Bulletin of the London Mathematical Society 42(5), p. 879-890, 2010. [Kir11] Jonathan Kirby, Finitely presented exponential fields. Algebra and Number Theory 7(4), p. 943-980, 2013. [Zil05] Boris Zilber, Pseduo-exponentiation on algebraically closed fields of zero. Annals of Pure and Applied Logic 132(1), p. 67-95, 2004. E-mail address: [email protected] URL: http://math.harvard.edu/~wboney/

Department of , Harvard University, Cambridge, Massachusetts, USA